- teacher config: bitrate 2.85 -> 4.8 kbps (128 dims x 3 bits x 12.5 Hz), quantizer is FSQ not Q2D2, PESQ 3.14 / ESTOI 0.874 on the 50-utterance set, note that only MRF kernels 3/7/11 are built
- sigreg config and model card: trained with a GAN discriminator; 50-utterance scores PESQ 2.33 / ESTOI 0.76
- main config: paper Table 1; no discriminator, no distillation loss
- model card: citation is @misc (APSIPA version was not accepted), multilingual SSL comparison stated correctly, unused EMA target and predictor weights noted
